Upload
ijartet
View
224
Download
0
Embed Size (px)
Citation preview
8/10/2019 Interpretation Based Building Extraction From Fusion Of LIDAR And Aerial Imagery
1/6
ISSN
2394-3777 (Prin t)
ISSN2394-3785 (Onlin e)
Avai lable onl ine atwww.ijartet.com
Internat ional Journal of Ad vanced Research Trends in Engineer ing and Technology (IJARTET)Vol. 1, Issu e 4, December 2014
29
Interpretation Based Building Extraction FromFusion Of LIDAR And Aerial Imagery
S.VELLAI SAMY1, S.BEEMA SABEENA
2
Professor, EEE, SCAD College of Engineering and Technology, Tirunelveli, India1
Student, M.E. VLSI design, SCAD College of Engineering And Technology, Tirunelveli, India2
AbstractBetter fusion of Light Detection and Ranging (LIDAR) and aerial imagery for building extraction is very much
needed application for continues updates for urban planning and survey information. This reason motivates to develop a
seamless fusion of LIDAR and aerial imagery on the basis of aspect graph driven method. The feature of houses such as
geometry shapes and structure is also explored. An aspect represents either projection or part of projection of 3D house
primitive. By using aerial image processing in combination with results of LIDAR data processing, hierarichal graph of aspectare constructed. In the aspect graph describe the node represents the face aspect and the arc is described by attributes obtained
by the formulated coding regulations, and the co-registration between the aspect and LiDAR data is implemented. As the result
, the aspects graphs are interpreted for the extraction of houses by using MATLAB.
Keywords: Aerial imagery, LIDAR, extraction, image processing
I. INTRODUCTION
Now a day, spatial information has become one the most
important information source in human life. Spatial
information is clearly expected that more information shall be
associated with survey information provided through system
such as GIS geo information system. Building extraction ismost important application used for continues updates for
urban planning, cartographic mapping, and civilian and
military emergency responses. The efforts for building
extraction using a variety data sources have made in imageprocessing, and computer vision communities. Light
Detection and Ranging is determine distance by using analysis
laser light to reflect mirror. This LIDAR data can be used for
mapping areas of building and application of industry and
aerial image is defined to the image from elevation to groundlidar has high resolution techniques it has high accuracy.It is
remote sensing technology that emits focused beams of light
and measures the time it takes for the reflections to be
detected by the sensor. This information is used to computeranges, or distances, to objects. In this manner, LIDAR is
analogous to radar (radio detecting and ranging), except that it
is based on discrete pulses of laser light. The three-
dimensional coordinates (e.g., x,y,z or latitude, longitude, and
elevation) of the target objects are computed from 1) the time
difference between the laser pulse being emitted and returned,
2) the angle at which the pulse was fired, and 3) the absolute
location of the sensor on or above the surface of the Earth.
Early methods can be grouped into three categories in terms of
the employed data sources. The first category uses optical
imagery, such as aerial imagery and high-resolution satellite
imagery. Building-detection method may not detect buildings
if the contrast between their rooftop and the background is
low. Since only using the gray scale information in the
satellite image, this is reasonable. Moreover, some closely
spaced buildings may not be detected correctly.This methodattempted to extract the boundary of a building to identify the
buildings, but separating building boundaries from non-
building objects, such as parking lot is difficult. It is especially
difficult to extract the complete boundaries of a building from
single optical image due to unavoidable occlusions, poorcontrast, and shadows. The second category utilized active
imagery data, such as Light Detection and Ranging (LiDAR)
data and radar imagery. The third category integrates both
high resolution optical and active imagery data. Thereforemany efforts have been made on the fusion of Light Detection
and Ranging (LiDAR) and aerial imagery for the extraction of
houses, on taking advantage of a buildings geometric
features, properties, and structures for assisting the further
fusion of the two types of data has been made . Applications
are urban city or town planning is the discipline of land use
planning which explores several aspects of the built and social
environment of municipalities and communities. Using fusionof LIDAR and aerial imagery on the basis of aspect graph
method to create digital building models therefore large area
8/10/2019 Interpretation Based Building Extraction From Fusion Of LIDAR And Aerial Imagery
2/6
ISSN
2394-3777 (Prin t)
ISSN2394-3785 (Onlin e)
Avai lable onl ine atwww.ijartet.com
Internat ional Journal of Ad vanced Research Trends in Engineer ing and Technology (IJARTET)Vol. 1, Issu e 4, December 2014
30
models to create very short space of time.Recently, the
methods of third category have been widely used the efficient
fusion of the two types of data sources can compensate for theweaknesses of one data to another. However, most current
efforts have still staying the study of the gray level fusion and
few efforts on utilization of the geometric or chromaticfeatures of buildings to support the deep data fusion for
building extraction. To confront these issues, develops an
aspect graph-driven method to explore how the features of
houses are utilized for the data fusion between aerial imagery
data and LiDAR data. The method attempted to extract the
boundary of building to identify the buildings and nonbuildings.
II. DESIGNGOAL
The proposed methodology consists of aerial image
processing, and principle of aspect code.
The block diagram of the proposed methodology is shown
in Fig. 1. The details of each phase are given below.
Fig. 1. Proposed method
1.AERIAL IMAGERY PRE PROCESSINGAim of image pre-processing is an improvement of the
image data that suppress undesired distortions or enhance
some image features relevant future processing and analysistask.There are following steps they are image enhancement
and noise removal, edge detection, edge grouping, Kalman
filter to track line.
Step 1- Image enhancement and noise removal:
The principal objective of image enhancement is to
process a given image so that the result is more suitable thanthe original image for a specific application It accentuates or
sharpens image features such as edges, boundaries, or contrast
to make a graphic display more helpful for display and
analysis. The enhancement doesn't increase the inherent
information content of the data, but it increases the dynamic
range of the chosen features so that they can be detected easilyas shown in Fig. 2a and 2b.Captured aerial image is
unavoidable for the image to involve noise. Considering is
used to accentuate edges. As shown in figure has used median
filter If the objective is to achieve noise reduction instead ofblurring, this method should be used. This method is
particularly effective when the noise pattern consists of strong,spike-like components and the characteristic to be preserved is
edge sharpness. For each input pixel f(x,y), we sort the values
of the pixel and its neighbours to determine their median and
assign its value to output pixel g(x,y). The typical high-pass
filter mask is
The original image can be converted to the enhanced image
Fig.2. (a) original image and (b) enhanced image
Step2-Edge detection
There are many ways to perform edge detection but
in proposed method canny method has used mask for edge
extraction fig 3. Directly this method for house extraction.When the mask are convolved with a moving window
strongest response is taken as edge and corresponding
orientation is taken as the direction of an one building the
mathematical model of output value of gray value can be
expressed as
8/10/2019 Interpretation Based Building Extraction From Fusion Of LIDAR And Aerial Imagery
3/6
ISSN
2394-3777 (Prin t)
ISSN2394-3785 (Onlin e)
Avai lable onl ine atwww.ijartet.com
Internat ional Journal of Ad vanced Research Trends in Engineer ing and Technology (IJARTET)Vol. 1, Issu e 4, December 2014
31
(2)
where f(m,n)is gray value of original image and w is the maskvalue are obtained. Using edge detection mask the gray values
of building edges are enhanced and the background noise is
refined efficiently.
Fig. 3 Result of extracted edges
Thus the candidate pixel of the building edges can be
extracted .after thinning of edges result can be extracted.
Step3-Edgegrouping:
A building boundary is segmented into several
segments after detecting the building edge pixel. For this
reason, the following task it to trace the edge fragment pixels
and then group them. The method to group two neighbour
edge fragments into one new edge fragment is proposed asfollows Fig. 4.1) The slopes of two neighbour edge fragments meet the
condition of | | slope threshold, where and represent
the slopes of two edge fragments, respectively.
2) The distance D between the ending point of an edge
fragment and the starting point of another edge fragment
should meet the condition of D dist_threshold.
Fig. 4. Merging neighbour edge fragments.
When the two conditions above are met, two neighbour edge
fragments are merged by
AD = AO1 + O1O2 + O2D. (3)
The same operation is above repeated for each edge fragments
until all edge fragments are finished. After this step, most of
the edge fragments are smoothly connected.
Step4-to Track line using Kalman fi lter:
Noisy data inHope fully less noisy data out
The above operations unavoidably cause the gaps along the
edges of a house. Kalman filter algorithm is proposed to track
the broken-off edges to fill the gaps. The basic idea of thismethod is that each edge is considered as a track of a moving
point on an edge. When the broken-off edge fragments are not
detected, the Kalman filter keeps track of the next fragment. Ifthere exist two edge fragments, which meet the two conditions
in Step 2, they will be merged into one edge fragment. When a
broken-off edge is detected, a number of the pixels are
prefilled. The filled number is required to assure that the
methods can the next edge fragment. This computational
process is mathematically described as follows: If an edge istracked, let X (k) be the state vector of movement, x (k) be the
moving location, x(k) be the velocity, y(k) be the observation
location that is the location of the detected edge point, T be
the step size, and k be the tracing time; the state equation ofconstant velocity of point D is
X ((k + 1)/k) = AX(k) (5)
Where X = [x x] T, and A = . The measurement model
is simplified as
Y (k) = CXT(k) (6)
Where X = [1 0] T, and Y(k) =[y y]. The predicated state
equationX ((k + 1)/k) = A^X (k). (7)
The output of the filter is
(8)
Fig.5. Using the formulated coding regulation the area, perimeter, and the
central coordinates of a face are calculated
After the Kalman filter in combination with step 2 is carried
out, an entire building boundary is extracted.
2. CREATION OF ASPECT GRAPH
With the detected edges above, the next task is to
create aspect and aspect graphs using the proposed method in
principle of aspect code and creation of aspect code database
8/10/2019 Interpretation Based Building Extraction From Fusion Of LIDAR And Aerial Imagery
4/6
ISSN
2394-3777 (Prin t)
ISSN2394-3785 (Onlin e)
Avai lable onl ine atwww.ijartet.com
Internat ional Journal of Ad vanced Research Trends in Engineer ing and Technology (IJARTET)Vol. 1, Issu e 4, December 2014
32
and then create a digital building model. To this end following
steps are proposed algorithm, each face is filled with gray
value, 255. An example shown in Fig.5
Face coding:
The first step of aspect creation is face coding to the
boundary detected in precious section. The following steps are
conducted.
Face fil ling:
Example with face filling, the area of each face is
obtained by accounting the number of the filled pixels. Afterfilled, a binary imagery is created. Procedures of coding a
house using the formulated coding regulation, in which the
small circles indicate an endpoint or the corner of a line or a
curve. The orientation and length of a straight segment arecalculated. The area, perimeter, and the central coordinates of
a face are calculated as well.
TABLE I
ATTRIBUTE VALUES OF NODES AND ARCS
Boundary detection:` With the binary imagery created above, a gray
threshold of 100 is setup to detect the boundary. After this,
thinning edges and deleting edge burrs are implemented
consequently. Finally, the boundaries with a pixel width are
obtained.
Boundary vectorization:With the result obtained above the boundary of each
face is vectorized using an eight neighbour tracking operator.
With the operation, the perimeter of each boundary can beobtained and is used as an attribute value of the face aspect.
Corner detection:With the vectorized boundary of each face above, the
corner detection is conducted using a method called maximum
curvature difference between forward K steps and backward K
steps.
Selecting one pixel along the boundary, we called
central pixel, and calculating the curvatures of the line located
in the selected pixel (noted k0), the forward 3 pixels (noted
k+3), and the backward 3 pixels (noted k3), respectively.
The curvature differences between the selected and the
forward pixels (noted k0+3) and between the selected and the
backward pixels (noted k03) are calculated. respectively.
The corner is detected by using the second curvature
difference, i.e.,Curvature0=|k03k0+3|
(9)
If Curvature0 is greater than the given threshold, thepixel is determined as a corner; otherwise, it is not a corner
pixel. Sometimes, it is necessary to suppress the local no
maximum since the multiple Curvature0s surrounding the
central pixel simultaneously meet the condition of (9).
Determination of straight line or curve: When using the
aspect or aspect graph for house interpretation, the property ofline segmentation, either a straight line or a curve, has to be
determined. To this end, the curvature of the line segmentationbetween two corners is calculated. If the curvature is close to
zero, the line segment is considered as a straight line,
otherwise, as a curve.
Face coding: With the operations above, face coding is
conducted in terms of the coding regulation formulated . When
the above operations are finished, the face codes are completed.
With the faces coded above, the aspect graph is
constructed. The basic steps include: First, each face is given
an attribute name using natural numbers 1,2,3..Meanwhile, the
attribute values are calculated using the parameters listed inTable I.
Second, the aspect graph is constructed using nodes
and arcs. Each face is taken as a node of the aspect graph, andeach of the shared edge is determined by the same imagery
coordinates and taken as an arc of the aspect graph [as shown
in Fig.6. Third, a primitive aspect graph is constructed using
the aspect merging regulation described in coding regulation
of aspect coding
Fig.6. Illustration of aspect and aspect graph construction, (a) the segmented
houses, and (b) and their aspect graphs.
For example, face aspects 5 and 6, face aspects 9 and
10, and face aspects 2 and 4 are merged in Fig.6(b). In
addition, face aspect 1 is an isolated aspect, which represents
an independent house. Fourth, the aspect graphs between the
created in this section and the archived in the database are
correlated. With this operation, the aspect graph using the
proposed face-coding regulation can effectively remove the
mis-segmented area .
Attribute values for Node Attributes values for Arc
1.Area
2.Perimeter
3.Central coordinates
4.Orientation angle of a
face aspect
5.Codes describing faceaspect
1.The coordinates of a
shared edge
2.The types connected
edges
3.The slope of straight line
4.The curvature of a curve5.The information of two
end points
8/10/2019 Interpretation Based Building Extraction From Fusion Of LIDAR And Aerial Imagery
5/6
ISSN
2394-3777 (Prin t)
ISSN2394-3785 (Onlin e)
Avai lable onl ine atwww.ijartet.com
Internat ional Journal of Ad vanced Research Trends in Engineer ing and Technology (IJARTET)Vol. 1, Issu e 4, December 2014
33
3. COMBI NATION BETWEEN TWO DATA
The created aspects above unavoidably contain theobjects that do not belong to houses, such as vehicles or
parking lots.
Fig. 7. LiDAR footprints.
For this reason, the application of elevationinformation driven by LiDAR footprints as shown in Fig.7.
will help separate ground objects and non-ground objects. A
common method is to co register the LiDAR data and the
created aspects, with which the houses boundaries and 3 -D
co-ordinates can be obtained and are used to further, created.
aX + bY + cZ = 1 (10)
where a, b, and c are unknown coefficients, and X, Y and Z
are coordinates of LiDAR footprints shot within the houses
roof.
Fig.8. Coarse 3-D models of the houses through co-registration between the
aspect and the LiDAR data.
As shown in Fig.8, at least three LiDAR footprints are
required to determine the equation coefficients a, b, and c by
least squares estimates. With this algorithm, a coarse 3-D
model of houses can be generated. As observed in theboundaries of the 3-D model of each house are very coarse
since it is difficult to extract the boundaries of each house
from single LiDAR data.
4. ASPECT INTERPRETATION BASED BUI LD ING
EXTRACTED
The primary design for the extraction of houses using
aspect interpretation is a correlation operation between theaspect codes created from aerial imagery (called image code)
and the aspect codes archived in the codebase (called
codebase). The correlation operation is carried out by theattribute values assigned in both the nodes and the arcs of the
aspect graphs. Each node and each composed of many
attribute values; a linear combination of these attributes is
constructed by
node attribute=Area + Perimeter + Central Co-ord +Orientation +Code aspect (11)
arcattribute=Co-ord Twoends+ Type edge +Slope
line+ Curvature curve (12)
The correlation coefficient maximum of the note/arcattributes between the obtained from aerial imageries and the
archived in the database is taken as criterion. With the new
combined variables in (11) and (12), the two criteria are
employed to determine their correlation, respectively. The first
criterion is the correlation coefficient maximum of the node,
which is employed for determining the candidate house undera given threshold of correlation coefficient. The successful
operation of this step will be able to interpret and find a
significant amount of geometric information of a house, such
as aspect property, geometry, gray area, relationship, and
separate the ground objects (e.g., parking lots) and non-groundobjects (e.g., houses). The second criterion is a correlation
coefficient maximum of the arcs between the obtained from
the aerial imagery and the archived in the database. The same
operation as the node correlation is carried out. The successful
operation of this step will be able to interpret and find a
primitive graph, which describes an entire projection or partialprojection of a 3-D primitive.
Fig. 9. Building and non building identified
If an aspect is identified as a house, we will further determine
how many LiDAR footprints hit its roof .Repetition of the
operations in Step 4 and Step 5 is required in order to avoid
8/10/2019 Interpretation Based Building Extraction From Fusion Of LIDAR And Aerial Imagery
6/6
ISSN
2394-3777 (Prin t)
ISSN2394-3785 (Onlin e)
Avai lable onl ine atwww.ijartet.com
Internat ional Journal of Ad vanced Research Trends in Engineer ing and Technology (IJARTET)Vol. 1, Issu e 4, December 2014
34
missing the extraction of houses. Fig.9 shown extraction of
buildings can identified when above process can be repeated .
III RESULTS AND DISCUSSION
Create codes in MATLAB main work of the method
includes following aerial image preprocessing including edge
detection using detector edge pixel grouping gap filling is
developed. Select the image primitives and compared data.
Aspect graph are interpreted for extraction of building. These
methods attempted to extract the boundary of a building from
non building objects as shown in fig 9. The proposed method
is to extract building.
IV CONCLUSION
In this work, The main work is input captured aerial
image can be pre processed including edge detection using a
detector mask, edge pixel grouping, and gap filling, isdeveloped. Types of 3-D house primitives are selected, and
their projections are represented by the aspects. Then coding
regulations are formulated for creation of the (hierarchical)
aspect graphs on the basis of the results of aerial image
processing and the results of LiDAR data.
In the aspect graphs, the nodes are represented by the
face aspect, and the arcs are described by the attributes using
the formulated coding regulations. Result aspect graph are
interpreted for the extraction of houses and the extracted
houses are fitted.
V FUTURE ENHANCEMENT
Future work is multiple target extraction from
satellite image using EM method TLCC technique aims at
overcoming this problem by also allowing the
characterization. The expectationmaximization (EM)
algorithm. Pixels belonging to the land-cover classes of
interest are identified by means of an iterative labeling
strategy based on Markov random fields segmentation model
can be developed by K-means based on multi level threshold.
Final TLCC are iteratively updated through an MRF-based
strategy. Therefore future developments are planned in these
two directions, as well as toward comparison with target
detection methods.
REFERENCES
1) Ahmadi.A, Salman.A , Valadan Zoej ,M. J, Ebadi .H, andMohammadzadeh.A,(2012), Automatic urban building
boundary extraction from high resolution aerial images
using an innovative model of active contours,Int. J. Appl.Earth Observ. Geoinf., vol. 12, no. 3, pp. 150157.
2)
Axelsson.P(1999),Processing of laser scanner dataAlgorithms and applications ISPRS J. Photogramm.
Remote Sens., vol. 54, no. 2/3, pp. 138147.
3) Baillard.C and Matre.H(1999)3-D reconstruction of urbanscenes from aerial stereo imagery: A focusing strategy,Comput. Vis. Image Understanding, vol. 76, no. 3, pp.244258.
4)
Biederman.I(1985), Human image understanding: Recent
research and a theory,Comput. Vis., Graph Image, vol. 32,no. 1, pp. 2973.
5) Fujii.K and Arikawa,T.(2002) Urban object reconstruction
using airborne laser elevation image and aerial image,IEEE Trans. Geosci. Remote Sens.,vol. 40, no. 10, pp.2234224.
6) Rottensteiner.F(2003)Automatic generation of high-
quality building models from LiDAR data,IEEE Comput.Graphics Appl., vol. 23, no. 6,pp. 4250.
7) Sirmacek.B and Unsalan.C(2009.)Urban-area andbuilding detection using SIFT keypoints and graph
theory,IEEE Trans. Geosci. Remote Sens., vol. 47, no. 4,pp. 11561167.
8) Yan.G, Zhou.G, and Li.C(2007),Automatic extraction ofpower lines from large-scale aerial images, IEEE Trans.
Geosci. Remote Sens. Lett., vol. 4, no. 3, pp. 387391.9) Yu.Y, Buckles. B.P, and Liu.X (2009),Residential
building reconstruction based on the data fusion of sparseLiDAR data and satellite imagery,in Proc. 5th ISVC, II,
vol. 5876 LNCS, Lecture Notes in Computer Science,pp.240251.